Intelligent Flexible Automation David Peters Chief Executive Officer Universal Robotics February 20-22, 2013 Orlando World Marriott Center Orlando, Florida USA
Trends in AI and Computing Power Convergence of Artificial Intelligence Capabilities Hardware (Computers) 1 Software (Artificial Intelligence) 2 2 nd Generation 57 63 Transistors 3 rd Generation 64 71 4 th Generation 72 Now Integrated circuits Microprocessors 1 st Generation 56 74 Initial artificial intelligence 2 nd Generation 75 87 Expert systems 3 rd Generation 88 Now AI for specific industries & problems 5 th Generation Now Future AI devices with massive parallel processing 4 th Generation Now Future Intelligence based on learning pattern of living beings Overlapping Technology Vectors Nvidia computing: X teraflop Universal Robotics intelligence: Neocortex [1] http://www.webopedia.com/didyouknow/hardware_software/2002/fivegenerations.asp [2] http://en.wikipedia.org/wiki/artificial_intelligence
Big Data It s all a matter of perspective Flexible intelligence requires handling lots of data, but Big is not big for algorithms and computers Data reduction examples: 80KB of data for individual face recognition Cartons: 20KB of data for unique carton/package recognition Single bar code: 3KB data for specific label information Volume: 12,500 U.S. large distribution centers (> 100K SQ FT) Throughput: 5M cartons/yr/dc @ 12,500 = 62.5B cartons/yr Data on every carton for a year = 780 TB nvidia Parallel processor Tesla Kepler 10 Process simple calculation on all 780 Terabytes in under 3 minutes!!
Algorithm that Mimic Learning Artificial Intelligence uses sensor input to learn Sensory Motor learning loop: (act sense react) Bottom-up design Hardware agnostic Simplifies complexity/chaos Improves process via operational insight BIG DATA reduced for comprehension It s the Way the Real World WorksTM\ Use: Both data analysis and automated control
3-D Vision Animals with stereo vision understand depth intuitively Disparity Algorithms need Cartesian coordinates x, y, z Point Clouds 3D coordinates on an object surface E.g. UR combined in real-time 4 point clouds for composite 3-D Resolution the distance between the points Vision analysis uses traditional operators blob, edge detection, matching, measuring Sensors Structured Light, Camera pairs (Stereopsis), Laser, Light Detection And Ranging (LIDAR) Processing time >500ms (human reaction time 250ms)
Motor Control Real-time kinematics, path planning and obstacle avoidance High speed interface Machine reacts to variations of task based on sensing Any type of actuation whatever is necessary for the job from this: to this:
Automating IntelligenceTM 1. 3-D Sensing to find randomly placed objects Spatial Vision Robotics uses sensors for data analysis Maps 3-D space Scalable 3-D precision by utilizing a range of sensors Provides accurate 3-D vision guidance and 3-D inspection 2. Motor Control to drive machines reactively Autonomy software automates robot programming Integrates kinematics, path planning, & obstacle avoidance 3. Intelligence to learn new tasks Neocortex learns how to handle never-seen-before objects New form of Artificial Intelligence Responds dynamically to change with real-time sensory input Uses memory to match what is known with what it is learning
Intelligent Flexible Automation Applications Random Pick of Difficult Objects with Inspection Deformable objects bags partially filled Semi-rigid objects rubber blocks Cosmetic bottles clear, metallic, odd shapes Random 3-D Inspection Package tracking & sorting - random objects & labels, locations Random Depalletization Unlimited quantity of boxes mixed pallet Varying location and orientation - 6 DOF
Contact Information David Peters Chief Executive Officer Universal Robotics, Inc. PO Box 171062 Nashville, Tennessee 37217 USA Phone: (615) 366-7281 davidpeters@universalrobotics.com www.universalrobotics.com